Vakonomic Principle

The principle of stationary action stands as one of the most elegant and powerful ideas in classical physics, stating that nature follows a path of least effort. This single concept beautifully derives the laws of motion for uninhibited systems. However, a profound challenge arises with nonholonomic constraints—rules that restrict an object's velocity, like those governing an ice skate or a rolling coin. Such constraints break the simple application of the action principle, creating a knowledge gap that requires a different approach. How do we determine the laws of motion when the path is constrained in such a complex way?

This article explores two competing formalisms that address this question. First, in "Principles and Mechanisms," we will delve into the standard, physically verified Lagrange-d'Alembert principle and contrast it with the radical vakonomic principle—a mathematically beautiful attempt to save the action principle. We will uncover how a subtle difference in their formulation leads to dramatically different physical predictions and geometric structures. Then, in "Applications and Interdisciplinary Connections," we will resolve this conflict by showing why the standard principle correctly describes mechanical systems, while the vakonomic principle finds its redemption as the fundamental theory for optimal control and robotics, ultimately revealing that two "wrong" answers can simply be solutions to two different questions.

At the heart of classical physics lies a principle of extraordinary elegance and power: the ​​principle of stationary action​​, often called Hamilton's principle. It proclaims that for a particle moving between two points in a given time, the actual path it follows is the one for which a special quantity, the ​​action​​, is stationary—typically a minimum. The action is an integral over time of the Lagrangian, $L = T - V$, which is the kinetic energy ($T$) minus the potential energy ($V$). Nature, it seems, is economical. It doesn't waste effort. From this single, beautiful idea, all of classical mechanics can be derived. But what happens when we are not free to roam? What happens when our motion is constrained?

Imagine a bead sliding on a wire or a train car fixed to a track. These are examples of ​​holonomic constraints​​. They are "nice" because they restrict the possible positions of an object. We can handle them easily by simply choosing a new set of coordinates that automatically respects the constraints (like the distance along the wire) and then applying the principle of stationary action in this reduced world. The beauty of the principle is preserved.

But some constraints are more cunning. Consider an ice skate on a frozen lake or a coin rolling on a table. The skate's blade can only move forward or backward, not sideways. The rolling coin must move in the direction it's pointing. These are ​​nonholonomic constraints​​. They restrict the possible velocities of an object, but not necessarily the final positions it can reach. You can skate from any point on the lake to any other, but at every instant, the direction of your movement is severely limited.

These nonholonomic constraints pose a profound challenge. You cannot simply reduce the number of coordinates, because the system can, eventually, reach any configuration. How, then, can we determine the equations of motion? Does the majestic principle of stationary action fail us here?

The traditional and physically verified answer to this puzzle is the ​​Lagrange-d'Alembert principle​​. It sidesteps the global, path-centric view of the action principle and instead adopts a local, instantaneous perspective. At any moment, imagine giving the system a tiny, instantaneous "nudge," a ​​virtual displacement​​, in any direction it's allowed to move. For the ice skate, this means a nudge forward or backward, but not sideways. The principle states that the forces of constraint—the forces that enforce the rules, like the normal force from the ice preventing the blade from slipping sideways—do no work during these virtual displacements.

This is a principle of "virtual work," not stationary action. It leads to a set of equations where the standard Euler-Lagrange expression, $\frac{d}{dt}\frac{\partial L}{\partial \dot q} - \frac{\partial L}{\partial q}$, is not zero. Instead, it is equal to a ​​constraint force​​, typically written as $A(q)^{T}\mu$, where the columns of $A(q)^T$ define the "forbidden" directions of motion and $\mu$ is a set of time-dependent multipliers determined by the dynamics. This approach accurately describes the motion of real-world nonholonomic systems. For time-independent systems, it also has the pleasing property of conserving mechanical energy. But for the purist, something may feel lost. We have abandoned the global elegance of the action principle for a local, differential rule.

This dissatisfaction leads us to a fascinating alternative: the ​​vakonomic principle​​, short for "variational axiomatic." The central idea is bold and simple: let's try to save Hamilton's principle. Instead of changing the principle, let's change the space of paths we apply it to.

The vakonomic recipe is as follows:

1. Consider the entire universe of possible paths a system could take.
2. Throw away all paths that violate the nonholonomic velocity constraints at any point in time. We are left only with paths that are "kinematically admissible."
3. Among this restricted set of admissible paths, find the one that makes the action $\int L \, dt$ stationary.

This is a beautiful and mathematically natural idea. It attempts to treat all constraints, holonomic or not, under the single, unifying umbrella of the principle of stationary action. Mathematically, this is achieved by constructing an ​​augmented Lagrangian​​, $L_{\text{v}} = L + \lambda^T (\text{constraint})$, where we introduce Lagrange multipliers $\lambda(t)$ that are themselves treated as new dynamical variables. We then find the stationary points of the action for this augmented Lagrangian, varying with respect to both the original coordinates $q$ and the new multiplier coordinates $\lambda$.

The difference between the Lagrange-d'Alembert and vakonomic principles boils down to a subtle but crucial distinction in what constitutes an "admissible variation".

• In the ​​nonholonomic​​ (Lagrange-d'Alembert) picture, we consider a valid path $q(t)$ that obeys the constraints. The variation $\delta q(t)$ is an instantaneous, virtual displacement at each point in time. We only require that the vector $\delta q(t)$ lie in the subspace of allowed velocities at that point. The varied path, $q(t) + \epsilon \delta q(t)$, does not itself need to be a valid path that obeys the velocity constraints. The variation is "virtual."

• In the ​​vakonomic​​ picture, the variation is of the entire path. We vary the path $q(t)$ to a new path $q_\varepsilon(t)$. The core demand is that this new path $q_\varepsilon(t)$ must also be a kinematically admissible path for all (small) $\varepsilon$. This is a much stricter condition. It means the variation is "tangent to the space of admissible curves". Linearizing the constraint $A(q)\dot{q}=0$ along the varied path leads to a more complex condition on the variation: not just $A(q)\delta q=0$, but $A(q)\delta \dot{q} + (\partial_q A) \dot{q} \delta q = 0$.

This seemingly small difference in the definition of a valid "wiggle" leads to dramatically different physical predictions. The vakonomic equations of motion contain extra terms, sometimes called ​​vakonomic forces​​, which involve derivatives of the constraint functions and, crucially, time derivatives of the Lagrange multipliers, $\dot{\lambda}$. These terms are entirely absent in the nonholonomic equations.

Let's put these two theories to the test with a classic example: a disk of radius $R$ rolling without slipping on a plane. The state can be described by the center's position $(x, y)$, its heading angle $\theta$, and its spin angle $\phi$. The "no-slip" condition provides two nonholonomic constraints relating these variables' velocities.

• ​​Nonholonomic Prediction:​​ Applying the Lagrange-d'Alembert principle, we find that the equation for the heading angle is $I_\theta \ddot{\theta}_{\text{nh}} = 0$. This means that if the disk is rolling straight, it will continue to roll straight. If it's turning at a constant rate, it will continue to turn at that constant rate. There is no spontaneous torque on the heading. This matches our physical intuition and experimental observation perfectly.

• ​​Vakonomic Prediction:​​ Applying the vakonomic principle yields a starkly different result. The equation for the heading angle is $I_\theta \ddot{\theta}_{\text{vak}} = R\dot{\phi}(\lambda_1\sin\theta - \lambda_2\cos\theta)$. This is not zero! The vakonomic equations predict a "spurious" torque that depends on the spin rate and the Lagrange multipliers. This torque would cause a disk rolling straight to spontaneously start turning. The difference between the two predictions, $\Delta = \ddot{\theta}_{\text{vak}} - \ddot{\theta}_{\text{nh}}$, is precisely this non-zero term.

The verdict from this and many other physical examples is clear: for nonholonomic systems, the Lagrange-d'Alembert principle describes reality, while the vakonomic principle does not.

The surprises of the vakonomic world don't stop there. Let's consider a simple system with no potential energy and time-independent constraints, for which we would absolutely expect the mechanical energy (the kinetic energy) to be conserved.

For nonholonomic dynamics, this is true. The constraint forces are always perpendicular to the velocity, so they do no work, and energy is conserved.

For vakonomic dynamics, however, this is not guaranteed! Consider a particle with the simple constraint $\dot{y} - bx = 0$. By deriving the vakonomic equations and calculating the rate of change of the kinetic energy $E = \frac{1}{2}m(\dot{x}^2 + \dot{y}^2)$, we find a shocking result:

This is not zero. The mechanical energy of the system changes over time, even though the Lagrangian and constraints have no explicit time dependence. The very act of enforcing the variational principle in this way has introduced a mechanism for energy to enter or leave the system, through the "power" supplied by the dynamically evolving multipliers. The mathematical elegance of the principle comes at the cost of a fundamental physical conservation law.

So, are these two principles forever at odds? No. They agree in one crucial case: when the constraints are ​​holonomic (or integrable)​​. A set of velocity constraints is integrable if, through mathematical manipulation, it can be expressed as a constraint on positions. If a constraint distribution $D$ is integrable, it means the system is effectively confined to move on a lower-dimensional submanifold within the larger configuration space, just like our bead on a wire.

In this case, and only in this case, the "extra" terms in the vakonomic force—the ones involving the curvature of the constraint distribution—vanish. The vakonomic and nonholonomic equations become identical. The discrepancy between the two formalisms is a direct consequence of the "non-integrability" of the constraints; it is a measure of how twisted the allowed directions of motion are. Integrability is the necessary and sufficient condition for the two dynamics to coincide for all mechanical systems.

These differing principles find a beautiful and deep expression in the language of modern geometry.

The standard Lagrange-d'Alembert dynamics are fundamentally tied to the ​​Riemannian metric​​ defined by the kinetic energy. The projection that separates the motion from the constraint forces is a metric-based orthogonal projection. The resulting flow is generally not Hamiltonian with respect to the canonical ​​symplectic structure​​ of the phase space. It fails to preserve this structure, a failure that is reflected in the non-vanishing of a mathematical object called an "almost-Poisson bracket's" Jacobi identity.

Vakonomic dynamics, on the other hand, is inherently ​​Hamiltonian​​. It can be perfectly described as a standard, albeit constrained, Hamiltonian system, but on an ​​extended phase space​​ that includes the multipliers $\lambda$ and their conjugate momenta as new coordinates.

So we are left with a fascinating dichotomy. Nonholonomic dynamics is physically correct but geometrically "messy," mixing metric and symplectic structures. Vakonomic dynamics is geometrically "clean" (purely Hamiltonian) but physically incorrect. This contrast reveals the deep and subtle relationship between variational principles, constraints, and the fundamental geometric structures that govern the laws of motion. The vakonomic principle, while not a descriptor of our physical world, serves as a brilliant theoretical foil, illuminating by its very difference the profound nature of the principles that do.

We find ourselves in a curious position. We have explored a beautiful mathematical structure, the vakonomic principle, which stems from the most elegant and powerful idea in theoretical physics: the principle of least action. It proposes that the trajectory of a constrained system is an extremal of the action, where the paths themselves are varied within the set of all motions that obey the constraints. It is a pure, unadulterated variational principle.

Yet, as we have seen, the equations of motion derived from this pristine principle often disagree with the equations from the standard Lagrange–d’Alembert approach, which we know from experiment correctly describes many constrained systems. This presents us with a wonderful puzzle. When two different but plausible theoretical ideas give two different answers, how does nature choose? And if one of them turns out to be "wrong" for describing a particular physical phenomenon, is it truly wrong, or is it perhaps the right answer to a different question? This is the journey we shall now embark upon, a journey that will take us from rolling coins to the heart of modern robotics and optimal control.

Let's start with a simple, familiar object: a coin or a disk rolling upright on a table. We all have a strong intuition about how it moves. If we give it a push, it rolls in a straight line. If it's already turning, it might trace out a large circle. This is precisely what the standard nonholonomic equations predict. Now, what does the vakonomic principle say?

It predicts something rather bizarre. For a disk set to roll in a perfect circle, the vakonomic equations predict that it won't stay in that circle. They predict a subtle, but real, change in the turning acceleration that would cause it to spiral away. Consider an even simpler case, the so-called "knife edge," which is like an ice skate constrained to move only forward or backward along its blade. If we imagine this skate moving in a circle at a constant speed and a constant rate of turn—a perfectly reasonable physical motion—we find a stark contradiction. The standard nonholonomic model handles this motion with no trouble. The vakonomic model, however, leads to inconsistencies; such a simple, steady turn is not a valid solution to its equations.

Nature, it seems, has spoken. For the everyday mechanics of rolling and sliding objects, the vakonomic principle appears to be wrong. The universe does not, in this case, choose the path that is a true extremal of the constrained action integral.

Why this failure? The answer lies buried in the mathematics. The vakonomic equations are exquisitely sensitive not just to the constraints themselves, but to how those constraints change as the system's configuration changes—a kind of "curvature" of the constraint space. This sensitivity introduces additional, phantom forces into the equations. In the case of a rotating rigid body with a constrained axis of rotation (the Suslov problem), these phantom forces manifest as gyroscopic-like terms that twist the body in ways that real physical forces do not. The standard nonholonomic principle, by its very construction, is immune to these mathematical subtleties and correctly captures only the true forces of constraint.

This physical failure hints at a deeper, more profound difference at the level of the mathematical structure of the theories. In physics, especially in classical and quantum mechanics, the most elegant and powerful theories are "Hamiltonian." This isn't just a name; it implies that the system's evolution occurs in a special kind of space—a phase space endowed with a "symplectic structure." You can think of this structure as a rule that governs how areas (or volumes, in higher dimensions) in phase space evolve. A key consequence, known as Liouville's theorem, is that for any Hamiltonian system, the volume of a cloud of initial conditions in phase space is perfectly conserved as it evolves in time. The cloud may stretch and deform in complicated ways, but its total volume remains invariant.

Here is the crux of the matter: vakonomic dynamics is, by its very construction, a perfectly Hamiltonian theory. It lives on a slightly enlarged or "augmented" phase space (which includes the Lagrange multipliers as variables), and on this space, it obeys all the beautiful rules of Hamiltonian mechanics, including the conservation of phase-space volume. Its underlying algebraic structure, the Poisson bracket, satisfies a crucial property called the Jacobi identity, which is the bedrock of Hamiltonian dynamics. The vakonomic world is, mathematically speaking, orderly and pristine.

The nonholonomic world, the one that correctly describes the rolling coin, is not so tidy. When we analyze its evolution in the natural phase space, we find that it is not Hamiltonian. It does not preserve phase-space volume. Its underlying algebraic bracket generally fails to satisfy the Jacobi identity. The degree to which it fails is directly related to the "curvature" of the constraints that caused the vakonomic equations to go astray.

This presents us with a fascinating philosophical choice. To describe the simple motion of a rolling disk, we must abandon the perfect, volume-preserving symmetry of Hamiltonian mechanics. Nature prefers a messier, non-Hamiltonian structure.

Could we see this difference in a laboratory? Absolutely. Imagine our rolling disk is driven by a small, periodic external torque. We can then observe its trajectory over many cycles and build a "Poincaré map," which is a snapshot of the system's state taken once every cycle. If the underlying dynamics were Hamiltonian (vakonomic), this map would have to preserve area. If we started with a small patch of initial conditions, the patch would deform, but its area after one cycle would be identical to its initial area. If, however, the dynamics are nonholonomic, the map would not preserve area. We would likely see the patch systematically shrink over time, indicating a flow that contracts phase-space volume. This provides a direct, experimental way to witness the profound geometric difference between the two models.

So, is the vakonomic principle a beautiful mathematical idea with no connection to reality? A failed theory? Far from it. It turns out it was simply the right tool applied to the wrong problem. To see its true power, we must shift our perspective from the field of mechanics to the field of ​​optimal control​​.

In mechanics, we ask, "Given these forces and constraints, how will the system move?" In optimal control, we ask, "Given these constraints, how should I steer the system to achieve a goal in the most efficient way possible?" Think about parallel parking a car. Your wheels cannot move sideways; this is a nonholonomic constraint, just like the rolling coin's. The question is not how the car will coast on its own, but what is the sequence of steering and driving inputs that will get you into the parking spot along the shortest possible path.

This is where the vakonomic principle finds its true calling. The trajectories predicted by the vakonomic equations are precisely the solutions to these kinds of optimal control problems. They represent the "straightest possible lines," or ​​sub-Riemannian geodesics​​, in a space where one's motion is constrained. The mathematical machinery of optimal control, particularly the celebrated Pontryagin Maximum Principle, when applied to finding the minimum-energy path for a constrained system, yields the very same equations as the vakonomic principle.

Suddenly, everything clicks into place. We do not have one "right" and one "wrong" theory. We have two distinct principles that answer two different, equally important questions:

• The ​​Lagrange–d’Alembert principle​​ tells us how a constrained system will naturally evolve according to the laws of mechanics. It describes the physics of a coasting bicycle or a rolling ball.

• The ​​Vakonomic principle​​ tells us the optimal path to steer a controlled, constrained system to minimize a cost, like time, distance, or energy. It describes the ideal path for a robot arm, a satellite maneuver, or a self-driving car.

This beautiful duality even has a point of reconciliation. If the constraints happen to be "integrable"—meaning they restrict the system to a lower-dimensional surface, like a bead on a wire—then the distinction vanishes. The path nature takes is the shortest path, and the nonholonomic and vakonomic equations become identical. It is only when the constraints are genuinely nonholonomic, allowing one to reach any configuration through clever maneuvering (like parallel parking), that the two principles diverge and reveal their distinct characters.

The story of the vakonomic principle is a wonderful lesson in the nature of scientific inquiry. A theory that fails as a description of one phenomenon can re-emerge as the cornerstone of another. It reminds us that the rich tapestry of mathematics is interwoven with the physical world in ways that are not always obvious, and a "wrong" answer is often just the right answer to a question we haven't yet thought to ask.